Cloudflare OS Glossary

A small-scale trial used to validate an idea.

Before full adoption, a small group runs one or more workflows to test impact and identify issues. For Cloudflare OS, we recommend a limited PoC that begins in a safe, primarily read-only scope.

See Getting Started and Setup for the six-step process.

A rollout that expands one department or workflow at a time instead of launching company-wide.

Start with one department and one workflow, then expand after establishing an operating pattern. This limits the impact of failure and reduces the burden on teams. Because Cloudflare OS is in Early Access, a phased rollout is more practical than a company-wide launch.

An approach that starts small and expands after proving value.

It is nearly synonymous with phased adoption. A typical example is starting with one read-only workflow.

Listing company workflows and identifying where AI may help.

This is the first step in a free workflow assessment. Work is prioritized by factors such as recurrence, documented decision criteria, and data location.

The value returned relative to the money invested.

A simple model subtracts adoption and operating costs from saved hours multiplied by labor cost. Because AI benefits can take time to appear, define how to measure them during the PoC.

A numeric target used to measure whether an initiative is working.

Examples include time to first reply and document-production time. KPIs are often used as PoC exit criteria.

The rules and controls that define who may do what.

For AI adoption, governance determines which data AI may see, who approves actions, and how records are retained. Cloudflare OS incorporates this through a zero-permission model.

See Security Design.

Monitoring behavior against rules and stopping deviations.

Governance defines the rules; control enforces them. This site uses identity and AI control to describe user management and AI-usage management together.

Support that helps new users become productive.

It includes training and initial setup assistance when employees start using a tool. Cloudflare OS support offers role-specific training.

A pilot that ends before reaching production.

The PoC may work but stall during security review, operational design, or organizational alignment. Avoid this by designing for production from the start of the PoC.

A model in which a person approves an action before execution.

Instead of executing AI decisions immediately, important actions wait for human review. In Cloudflare OS, Gatekeeper brokers these approval-gated writes.

A business platform for running AI agents safely inside a company.

Despite its name, it is not an operating system such as Windows or Linux. Cloudflare released it in August 2026 as an open-source enterprise AI agent platform.

See What Is Cloudflare OS?.

The browser environment where employees work with AI.

Agents write code and generate documents, slides, and business apps in isolated execution environments. Employees sign in to use it.

A control point between AI and internal systems.

When AI reads or writes internal data, Gatekeeper allows, denies, or routes each action for human approval and records access. It avoids giving API keys directly to AI.

See What Is Gatekeeper?.

A standard connection point between internal apps and AI.

MCP stands for Model Context Protocol, a standard way for AI to use internal systems. MCP Server Portal catalogs and manages those connections for an organization.

Cloudflare's serverless execution environment.

Cloudflare OS features run on Workers. A free tier exists, but enterprise use often requires Workers Paid, starting at $5 per month.

See Pricing and Cost Guide.

An authentication layer that limits access to employees.

Cloudflare Access integrates with the employee identity provider to control workspace login. Zero Trust is the security principle of trusting nothing by default and verifying every request.

A gateway through which all AI requests pass.

AI Gateway centrally tracks model usage, makes costs visible, and applies budget and rate limits. It also provides usage by department.

Blueprints for apps and workflows, plus their working components.

They let teams share and reuse effective approaches. A Blueprint is the design, and a Gadget is a component.

Background information for AI and definitions of what it can do.

Context supplies company-specific rules and assumptions; Skills define available capabilities. The industry scenarios use these terms.

See Use Cases by Industry.

Starting with no permissions and enabling only what is needed.

Every agent and app starts with zero access and can perform only actions approved by an administrator. It is the opposite of allowing everything first and restricting it later.

A limited release stage before general availability.

Cloudflare OS entered this stage after its August 2026 release. Because specifications may change, begin with a limited PoC and phased rollout rather than a company-wide launch.

A delivery model operated by Cloudflare.

The current model deploys to the customer's Cloudflare account. A managed version is planned for the Dashboard, but pricing has not been announced.

An interface through which systems communicate, and the credential used to access it.

API stands for Application Programming Interface. An API key is a credential for using that interface. Gatekeeper brokers access because giving the key directly to AI creates a misuse risk.

A standard for delegating access without sharing a password.

It powers experiences such as Sign in with Google. Gatekeeper uses it to specify which resources and actions AI may access.

The source of employee identities.

Examples include Google Workspace and Microsoft Entra ID, which centrally manage employee accounts. Cloudflare Access integrates with an IdP for authentication.

Using multiple services after signing in once.

Employees do not need separate passwords for each service, and accounts can be disabled centrally when someone leaves. SSO is implemented with an IdP.

Controls that prevent restricted information from leaving the organization.

DLP detects and blocks personal or confidential information leaving through email or cloud services. In Cloudflare OS, it is designed alongside AI Gateway.

A record of who did what and when.

Recording the data AI read and the actions it performed makes incidents traceable. Gatekeeper provides this record in Cloudflare OS.

Software used over the internet.

SaaS stands for Software as a Service; Gmail and Slack are common examples. Cloudflare OS is an open-source platform deployed to your account rather than a SaaS product.

See Product Comparison.

A license under which source code is public and commercial use is permitted without a license fee.

Apache 2.0 is an open-source license. The software license costs $0, but infrastructure and operations still incur costs.

Putting software on a server and making it available for use.

Cloudflare OS is deployed to the company's Cloudflare account.

See Deployment Guide.

The environment where a program actually runs.

In Cloudflare OS, agents write code in isolated execution environments. This prevents AI from directly accessing internal systems.

A pricing model that charges according to usage.

Workers and AI inference use this model. Costs may start low but grow with usage, so AI Gateway manages budgets.

A system central to company operations.

Examples include order management, inventory, and accounting systems whose failure can stop the business. They usually require custom connection design beyond standard integrations.

Knowledge and practical expertise accumulated inside a company.

Examples include past support answers, procedures, and configuration notes. AI can provide better answers when it can reference this material.

A business process expressed as an executable sequence.

A workflow can automate steps such as notifying an approver when a request arrives and routing it onward after approval. Cloudflare OS can create workflows from conversation.

A screen that presents metrics and status at a glance.

Dashboards summarize sales, support volume, AI usage, and other information in charts and tables. Cloudflare OS can generate them from conversation.

Your company's isolated area within a service.

In SaaS products such as Microsoft 365, a tenant is the area containing company data. Permissions are managed at this level.

An attack that overrides AI instructions with malicious content.

An attacker embeds text such as ignore previous instructions in data to trigger unintended behavior. Read-oriented operations and action restrictions in Gatekeeper reduce the risk.

Connections available without custom development.

This includes supported services such as Gmail, Slack, calendars, and documents. Core and payment systems require additional connection design.

Designing a custom connection where standard integrations are insufficient.

This is required for systems without a published connector, such as core systems, payments, or hotel property-management systems. It includes Gatekeeper configuration and sometimes MCP server development.

A three-stage approach to AI adoption.

Begin with read operations such as aggregation and summarization, then create documents and apps, and finally act by writing to systems. Opening permissions gradually limits risk.

Start with reads; require human approval for writes.

During the PoC, limit AI to reading internal systems. When write access is needed, route actions through Gatekeeper for human approval.

User management and AI-usage management considered together.

Access and IdP integration control who may use the system, while AI Gateway tracks how much they use. This is the third adoption step.

A Japanese grant that may cover up to 75% of training costs.

It supports employee reskilling costs for small and medium-sized businesses and may include a wage subsidy of ¥1,000 per training hour. It is a temporary measure through the end of fiscal 2026 in Japan, March 2027.

Check the latest information on the Ministry of Health, Labour and Welfare page.

A Japanese subsidy for introducing AI tools and related systems.

It targets small and medium-sized businesses and small operators, with up to ¥4.5 million available when four or more business processes are included. Applicants must work with a registered IT implementation support provider.

Check the latest information on the official site.

A registered support company required for this subsidy application.

Applicants for the Digitalization and AI Adoption Subsidy must apply with this provider. This service is registered as one.

The formal decision approving a subsidy payment.

Orders or contracts placed before this decision are ineligible. Apply with enough time in the schedule.

The number of business processes included in the subsidy scope.

The Digitalization and AI Adoption Subsidy covers implementations that improve four or more processes, such as customer inquiries, document creation, and data aggregation.

A training-plan filing submitted before applying for the grant.

For the Human Resources Development Support Grant, this notice is due at least one month before training starts. Prepare early because incomplete filings may be rejected.